Computational Assessment of the Regulation-Modulating Potential for Noncoding Variants

Abstract

26Large-scale genome-wide association and expression quantitative trait loci studies have identified 27 multiple noncoding variants associated with genetic diseases via affecting gene expression. 28However, effectively and efficiently pinpointing causal variants remains a serious challenge. Here, 29we developed CARMEN, a novel algorithm to identify functional noncoding expression-30 modulating variants. Multiple evaluations demonstrated CARMEN's superior performance over 31 state-of-the-art tools. Its higher sensitivity and low false discovery rate enable CARMEN to identify 32 multiple causal expression-modulating variants that other tools simply missed. Meanwhile, 33 benefitting from extensive annotations generated, CARMEN provides mechanism hints on 34 predicted expression-modulating variants, enabling effectively characterizing functional variants 35 involved in gene expression and disease-related phenotypes. CARMEN scales well with the 36 massive datasets and is available online as a Web server at http://carmen.gao-lab.org. 37 38

REVA as a Well-curated Database for Human Expression-modulating Variants

“…We used 2403 trained CNNs to annotate the functional effects of expression-modulating variants [26]. Most of the trained CNN networks were accurate, with an average AUROC of 0.908 and an average AUPRC of 0.904.…”

“…By the end of November 2019, REVA consisted of 11,862,367 entries covering 5,948,789 experimentally tested noncoding loci across 18 cell cultures from 14 publications [4,8,11,[29][30][31][32][33][34][35][36][37][38][39]. We first excluded loci tested in mice (15,152) and with more than one alternative allele (26,326 (Figure 2A, Supplementary Table S2). We found that both positive and negative variants were unevenly distributed on chromosomes, and no variants were located on the Y chromosome ( Figure 2B, Supplementary Table S3).…”

“…In efforts to pinpoint plausible regulatory mechanisms for these variants, we used 2403 trained CNNs to annotate the functional effects of sequence variations [26] based on 1249 TF binding, 766 histone modification, 280 DNA accessibility, and 108 DNA methylation profiles from recent Encyclopedia of DNA Elements (ENCODE) data.…”